Hi.  When you say "saved with the URL as the key" I'd like to understand your 
observation a little better.

When mod_pagespeed gets a resource from its HTTP fetcher, it stores it by the 
URL in the file-cache.  Depending on the optimization, we may also save the 
optimized results using the rewritten ".pagespeed." URL in the file-cache.

However, when we get a resource directly from the file system, we should not be 
caching it again at all; since it's on the file system already there is not 
much benefit.  We have unit-tests that count cache insertions with LoadFromFile 
and prove we don't re-store the input, but it's possible there's a corner case 
we've missed.

Exactly what are you observing in the cache?

I'm observing this problem incidentally and am seeing it reflected in my 
pagespeed cache dir.

Incidentally, let's say I pagespeed-map www.example.com/(user1|user2)/static to 
/var/www/static and that the HTML in those user dirs reference assets in the 
mapped static dir. When I hit www.example.com/user1, the first view is not 
pagespeed-optimized (as expected). A reload of the page a second later 
indicates that the async processing finished and that the optimized resources 
are now being linked to. Then, on my initial view of www.example.com/user2, I 
expected to see that the static resources on that page were referencing the 
already-optimized versions; they weren't.

What I dug into my cache to see if the resources were at least sharing the same 
cache entries, I found out that they weren't. I saw distinct entries like this 
for example:

/var/www/mod_pagespeed/cache/http,3A/,2Fwww.example.com/user1/static/css/W.test.
css,2Cqver=3.4.2.pagespeed.cf.CvyrCOB-uI.css,
/var/www/mod_pagespeed/cache/http,3A/,2Fwww.example.com/user2/static/css/W.test.
css,2Cqver=3.4.2.pagespeed.cf.CvyrCOB-uI.css,

In other words, the optimized test.css that's served from user1's and user2's 
HTML is optimized and saved to -- and I assume served from -- two different 
cache entries due to the URL being the cache key. I'd expect that there should 
only be one entry for the single optimized test.css, not two (and, in the case 
of thousands of users, I assume thousands of entries).

Issue 746 has been merged into this issue.

Clarification has been given.  We should work on improving our cache behavior 
in a few dimensions and this is one.

I would also like to see cache aggregation in the cases where the same domain 
is used for HTTP and HTTPS.