Most important: I think that the specific qualities of a particular startup or particular lab outweigh the importance of the category "startup" or "lab". Good labs, bad labs; good startups, bad startups.
You can often leave a bad startup faster than you can leave a bad lab.
Anecdote: I'm in a lab now and I'm really glad I'm doing what I'm doing. I like it a lot. It has the feel of a startup, but I'm doing research.
Much more but must go to lab! :)
Here are my thoughts:
1. If you're in a hot research area--the feeling of being a part of something bigger than yourself and the potential for impact is huge.
2. If you're in a hot research area and you're good--FYI good doesn't mean PhD... are you smart and do you get things done? Bringing ideas to the table helps too. If you fit these qualifications you'll soar to leadership in your area very quickly. I saw this happen with contractors, government program managers, and military personnel. I should note, the bench scientist part of a research organization is mostly flat so soaring to leadership is usually an informal designation. Still, its nice to see your ideas living and other people working on them either way.
3. Again, if you're in a hot research area (and the passion is there), then you'll get to work with cutting edge concepts and live there. Installing an SVN dump of open source package X is not cutting edge after working in research. At the same time researchers are much like the open source community, a big bulk tends to go in the same direction at the same time until the horse is shot, beaten, and then the bones picked clean.
4. Beware of the bureaucracy in a research environment. It is soul crushing and it will get you. It can really get in the way of everything you do.
5. The research world does offer you the opportunity to be entrepreneurial. Yes, really! Do you think you're good? Do you want your ideas listened to? Then go write a proposal and get funding. Getting someone to fund your ideas is incredibly validating (as I'm sure startup folks here will attest to). If you get enough funding, you can then afford to have people working on your team moving your idea forward. Maybe you start with a little money and then prove your concept enough to justify more funding later. Of course making relationships with parties interested in using your research helps too. Sounds a bit like growing a business doesn't it?
6. Oh last thing about research world--if you're willing to lead (if you're willing to start a company, you're willing to lead) then you won't just sit in a cubicle writing code. You get to write proposals, papers, reports. You get to go speak. You get to travel and reach out to other organizations. You also get to act as door to door salesman, I once went to Korea knocking on doors at US Forces Korea HQ peddling research. The number of hats you can wear is amazing.
I don't advocate doing one or the other. I do know if I make it in the startup world, I'll probably continue to spin multiple tops on the side, but I'd like to stay in the research world almost as a hobby. If one could eliminate the bureaucracy then it would be nearly a perfect job.
There is a big difference between being someone who starts a startup and working as employee #72 in said startup.
Likewise in the research world there is a big difference between someone conceiving ideas, trying to get funding, executing on them, and trying to transition the research (sales). Compare this to enthusiastic computer programmer who just wants to code and is unwilling to take on those other roles. YMMV
Then again, in professional sports, there is a big difference between being a player on the team and selling tickets at the door. Both are part of the same organization and both have a role to play.
In either environment you have to fight to matter but once you've had some success, then you get the freedom to make things happen.
In a research-heavy institution, it can be fun, but the scientists get most of the glory and creative work while the engineers just put the tools together for the customer or scientists. Research money tends to be less than startup money if the startup prospers.
Nobody can guarentee the startup will be fast paced, if it's indeed a startup.
This point isn't repeated enough. I've worked with startups where the pace was crazy-fast: like, finish that impossible feature that 3 other devs have tried and failed at by tomorrow, because we've got a demo. I've also worked at variants of this where the pace is the same, but the direction ends up changing every 2 weeks, so you never actually get anything substantive done. I've worked at startups where you work really fast, in one focused direction, for a long time and then find out it was the wrong direction and jump off a cliff. I've worked at startups that were really fast in the beginning, and then bogged down as we discovered the problem was much harder than expected. And then I've worked at startups where the pace was the same or slower than a normal company.
Startups are like a box of chocolates: ya never know what ya gonna get. ;-)
Yes these are not absolute rules, but definite observations.
The rewards are also different: does publishing a paper float your boat or do you want to change the world? Again, academia does produce great companies and products but it's a slower process.
-Norman Borlaug invented Dwarf Wheat while at a lab in Mexico. It has been estimated that this one invention saved well over a billion lives.
-The transistor. Depending on who you believe, it was either invented in an Austrian Lab, a German Lab, or at Bell Labs. Either way, it has revolutionized the world.
-The Internet. Came out of a combination of labs (defense contractors, CERN, etc)
-Penicillin. Discovered in a lab, and easily saved 100's of millions of lives.
The list could go on for another few hundred items pretty easily too. Do you seriously think that Startups help more people?
Of course, the potential for work in a research lab is huge too. So I think it's more about the specific opportunities. A web2.0 tagging social network is unlikely to change anyone's life right now, but startups like Kiva or 23andme might.