I 99 I started leasing a dedicated server - 200mhz with 128 megs for $100/month with decent bandwidth and transfer for my needs (10 gig/month to start with). I was able to put together quite a lot of projects and host them for that $99/month.
So, yes, undoubtedly things got cheaper, but not everyone was having to shell out $50k just to get any website up. I actually did a number of projects that would sometimes get sustained traffic (from usatoday, yahoo finance, etc) and had little problem on hardware that was costing perhaps $500/month.
Yes, it was the era of large VC funds pouring money in to lots of hardware/licensing - of course some people paid because it wasn't their money in the first place.
During the same period, I was also working at a company that had several multi-million dollar sites in development - massive budgets, hardware, project teams, etc. Heady times for sure. I got to see both sides of the fence at roughly the same time. Clients that had the millions would go to the agencies or hire top talent. Clients that didn't would still get their stuff done - it wasn't as 'all or nothing' as people like to remember.
MySQL was a viable option for us -- in fact, it beat Oracle in just about every benchmark we could dream of. Sure, it didn't have referential integrity, but we didn't need it.
Also, we were using Apache with a few scripts, no need for vignette or any other weird stuff.
Things are getting cheaper, but they were always reasonably cheap if you knew what you were doing, AND didn't try to follow buzzwords or have full ACID properties for your instant messaging status updates.
In 1998, you pretty much had to use commercial software if your site had any kind of traffic. That usually meant an Oracle database, and Oracle would only really scale vertically, so you had to spend even more up front on hardware that was powerful enough to scale.
Implementing Oracle on Sun wasn't something that most people (even good programmers) could do themselves, so they hired consulting companies, which cost even more.
Now, you might pay $5-10K to outsource an iPhone app. Then, you'd easily pay $100K just to get your database implemented, and easily that again for licence fees for your web software (Vignette anyone?).
Now, you pay $20/month for a basic EC2 setup, knowing you can scale it out horizontally. Back then you'd pay $50K up front for Sun hardware that would support the traffic you might get.
I don't know anything about the example given (DealPilot.com), but I do know that now days something that could run on shared hosting in 1997 could be hosted for close to free  now. That's still more than 10x less that $100/month.
I would also point out that developing an app for the iPhone costs $99 if you do it yourself (which appears to be how they built DealPilot.com), so that is still more than 10x less than $100/month over a year.
Lets compare costs for a year:
1997: $100/month = $1200, plus $3000 for server = $4200 (I assume there were hosting costs for the server, but I'll assume they were similar to the $100/month for shared hosting)
2011: $20/month for EC2 hosting (being conservative here - EC2 would let the company start off a lot cheaper than this and scale expenses as revenue grew) + $99 Apple fee = $339
 Free hosting examples include: Google AppEngine up to its free tier, Amazon's free EC2 offer. There are numerous examples of < $5/month virtual servers available, too.
If something takes off I'll buy a dedicated server for it, but if you're starting a company, get a Dreamhost account for $10/month and they throw the domain in for free. You can buy yourself two years of hosting runway for $200. If you max your dreamhost account you're well on your way.
I don't even bother paying for github, just throw git on my dreamhost account. (They have dedicated SVN services as well)
Also, I just got about a dozen notifications today that dreamhost had upgraded all my wordpress sites to 3.0.4. I don't do anything they take care of it. Someone breached an account? I get an email.
The only time I've ever run into trouble with dreamhost is an old old e-commerce site that needed MySQL4. I could get the ancient version of PHP it used working but couldn't get MySQL4.
It might not be the fastest, but you can't beat the value, especially if someone you know can give you a discount code (30OFFER is mine. I think I get some money off my bill if anyone uses it.. )
I glommed space on that machine for my own stuff for free - but I still had to pay $20 a month to host the DNS!!
But I guess that the old Sparcstation could only handle a tiny load. Scaling it up to thousands of customers would have been very expensive.
But yes - if we'd had to, it would have gotten expensive pretty quick.
Also, lots of web startups then couldn't get by on a shared-hosting account or a used Sparc 20 in a colo, because it cost so much more to render a dynamic web page or to store data. And lots of people were paying Solaris licenses, Oracle licenses (MySQL didn't exist, and Postgres was still pretty flaky and slow), Netscape server licenses, and didn't F5 start selling BigIPs for load-balancing about that time? And your domain name cost $100 a year.
There are so many vectors for comparison, it's ridiculous to try to generalize it. Seriously, different startups today cost several orders of magnitude different amounts today based on what they are doing.
To make any kind of reasonable comparison, you have to look at specific costs. Sure, processing, memory and bandwidth are way more than 10x cheaper, but applications might eat up a lot of that with additional complexity. Open source software, modern browsers, the rise of REST APIs, and the relatively saturated and savvy userbase mean that some apps today are infinitely better than they were 12 years ago (because they were previously impossible). On the other hand, people costs aren't any cheaper, the optimal size of a dev team hasn't changed, and you'll still need to expend the same amount of effort to compete regardless of how much better all technology products are overall.
Something else we also tend to forget is that in 1998 the mere act of writing HTML was an overpriced and highly sought after skill in the hands of a very few. These days anyone from a child to a great grand parent can author a webpage without thinking about it by using Facebook.
In the good old days, (1998), no one used Java on the server side. Java Applets were still not dead, and Java on the server was still not really born.
Apache SSI and perl were kings. AOLServer+tcl was probably the best available (though only known and used by a small group).
Java? We're not talking about the same 1998
I think Matt's Script Archive was the face of that popularity. It still exists today http://www.scriptarchive.com/
Sure, today you can develop prototype or MVP at fraction of the costs in 1998, but it's only 10% of total costs. The 90% is maintenance and support - and today it's not cheaper than then.
It can only be cheaper if founders willing to work for free. But this doesn't scale - sooner or later you need to hire employees, which will be demanding market priced salaries.
So, why then angels and VCs pushing entrepreneurs to do lean startups? The answer is simple: it allows them to reduce the risks, at the expense of entrepreneurs.
It's true that eventually a growing business will take on people who demand nothing more than market-priced salaries, but the people you need to get off the ground won't be satisfied with that.
VCs pushed them hard to get users at any cost and ipo, since you could do so easily without profits.