We call GetConnectionUnixProcessID and GetConnectionUnixUser *a lot*.
And we do so synchronously. Both is a problem.

To avoid the first problem, cache the last few requests with each cached
value being valid for one second.

On a quick test, this saves 98% of the requests:

     59 GetConnectionUnixProcessID(*)
   3201 GetConnectionUnixProcessID(*) (served from cache)
     59 GetConnectionUnixUser(*)
   3201 GetConnectionUnixUser(*) (served from cache)

Note that now as we serve requests from the cache, it might be the case
that the D-Bus endpoint already disconnected. Previously, the request would
have failed but now we return the cached user-id and process-id. This
problem is mitigated by only caching the values for up to one second.
Also, it's not really a problem because we cache sender names. Those
are supposed to be unique and not repeat. So, even if the peer already
disconnected, it is still true that the corresponding PID/UID was as
we have cached it. We don't use this API for checking whether the peer
is still connected, but what UID/PID it has/had. That answer is still
correct for the cached value after the peer disconnected.