At my employer we send out homework exercises, and I personally did the backend developer exercise before we sent it to anyone. I did this specifically to test how long it took. (For the frontend exercise, we didn't have anyone skilled enough on staff to do it, which is why we were hiring a frontend dev).
(1) It may seem counterintuitive to some, but my own general policy is, when it comes to little stuff ("how long did it take you to do X"), you just have to trust people, to a certain extent. Sure, some people may blatantly or grossly lie. But most likely these people will reveal their slipperiness in other ways, very very quickly.
Meanwhile -- and I think people are quick to overlook this -- playing the "policeman" role in every transaction with the candidate brings substantial negatives. By definition, it's adversarial. And generally there are (nearly always) non-adversarial ways to get the same information about the candidate ("Are they basically honest?") you're looking for. They take creativity (and an ability to read emotions and pick up on other signals), but they're there.
(2) Much bigger -- really, there's no need to sweat about the time to completion at all. Just look at the quality of the code.
It all just comes down to the fact that everything is interconnected: Good people generally turn out good stuff in reasonable amounts of time. When you're looking at good code, there is, I find, an intrinsic aura of ease and comfort which shines through it -- such that you just can't imagine it took them very long to produce it. Everything just flows -- just like it does when you talk to them.
Mediocre (and dishonest) people, on the other hand... never produce good stuff in virtually any amount of time. Sure, they can take the whole weekend to polish off their code... but it will still look bad, or at best, "Meh".
There might be some false positives (or outright frauds) by this approach), but I suspect very few. And those that do slip through, are easy to spot by other means (such as asking them to talk about their solution, for even a couple of seconds).
As to how likely it is that people will lie, given that we've hired several people who did this homework and they've proven to be as competent as we believed, there's at least some anecdotal evidence that some people have not lied.
You have to ask someone with no skin in the game.
Beyond that, I think "a few hours" is a bit too much to ask for, especially since you are presumably following up with an in person interview centered around the assignment. That's a big time commitment, and the commitment on the assignment especially is very asymmetric.
I'm not a big fan of assignments for this reason. They are just so asymmetric. I've been given interview homework that was supposed to take "about an hour" and it was more like 4 to 5 in reality. It felt like an unreasonable time request. I did the assignment, and did in fact tell them it took considerably longer than their resonates. I was invited for an on-site after all that, but declined for other reasons.
It's how the universe works, basically.
Isn't asking people to fly out for 6-8 hours of interviews (which effectively takes 3 days minimum out of your life) a much greater burden than spending 2-3 hours on some coding which you can do at a time of your choosing? (Followed by 2-3 hours of video chat interviews spread across multiple days, no flying required).
Also, just to clarify, the homework problem is _not_ something related to our business. The solutions are of no business value to the company.
I am not saying that you are wrong, but applying to Google does sound like it will have many benefits above your average company.
As to how much we pay ... Our pay is very good, and all but one of the positions for which we've had the homework requirement have been telecommuting as well. It's actually a pretty desirable place to work if you like good pay, telecommuting, working at a small company, a low pressure environment (we've been profitable for many years), and various other perks (training budget, flexible hours, blah blah blah).
There were no other limitations put.. "bonus points for stream based input" "bonus points for test cases" ... only a couple people actually delivered a "working" solution (one that ran and outputted anything), none of which had the correct output (one was close enough), and none had any tests.
It was truly something that should take a "skilled" developer a couple hours. And not something that should be entirely alien. To say the least, I was really disappointed with the results. I did the project myself in about 2 hrs, with test cases, 100% code coverage. (before I even gave it out, it was as trivial a challenge as I could come up with for a real world problem).
Why should I have to pay the couple dozen people for their 3 hours, when none delivered a correct solution?
In the end, the person with the closest to correct solution, was the one with the least experience... that person got the job.
And I had someone on my previous team make a point of bringing up this fine point at a meeting, also ;)