"I've been looking at some other high traffic sites and no one else seems to be doing this"
Really? That surprises me quite a bit... it's kind of considered step one for decreasing page load times. See here: http://developer.yahoo.com/performance/rules.html. If you want to test adherence to load time best practices, you may also want to try their performance testing tool YSlow: http://developer.yahoo.com/yslow/.
Here's why this is considered a best practice: browsers have a limit of around 8 simultaneous connections, and a limit of 2 (more for non-conforming browsers) per specific server. So if you're producing multiple files for a given page then they will load somewhat serially, since your browser will make fifteen round-trips to your server, two at a time.
So let's see how this plays out. Suppose you've got fifteen files at 10k each, or you could munge them together into a 150k file. Say the latency to the server is 65ms, and the download rate is a measly 25k/sec. Let's do the math... here's the equation:
$page_load_time = ( ($simultaneous_connections * $load_time_per_file)
+ $latency_per_request )
* ($number_of_files/$request_cycles)
Where:
$request_cycles = max(1, ($number_of_files/$simultaneous_connections))
So, with 15 files at 10k each:
((2 * .4 seconds) + .065 seconds) * 7.5 = 6.4875 seconds
or with a single 150k file:
((1 * 6 seconds) + 0.065 seconds) * 1 = 6.065 seconds
Not a dramatic difference, but still almost half a second of user waiting time!
Now - what happens when they're on a 3g connection? Or having a little wi-fi trouble? Let's try more bandwidth (100k/sec) and more latency (250ms):
((2 * .1 seconds) + 0.250 seconds) * 7.5 = 3.375 seconds
or
((1 * 1.5 seconds) + 0.250 seconds) * 1 = 1.750 seconds
Woah! Now all those separate little files has almost doubled the page load time!
Add in a whole bunch of other files, your fancy TypeKit load, ad server load, Facebook plugins, etc. etc. etc. and this stuff starts to matter even more.
GET /whatever.css ... If-Modified-Since:per page load, but it will be negligible in terms of practical gain, I suspect. Even then, I can't see that memcache would help much for a single static file, plus every time a new client arrives/the cached copy expires, you have to transfer the whole lot again. In general I'm on your side with this, what they are suggesting sounds like a silly idea to me.