It sounds as though the compiled binaries optimize the compute portion, which the interpreted php code was taking 10 secs to run. It's still likely running as a cgi process, that is as a fork/exec in a separate address space. If you're using mod_php or one of the many php engines available for modern web servers, you're already enjoying a scalability and execution time improvement over standard cgi. Except with compute code, apparently.

Here is the C library that I have used in the past:

https://sourceforge.net/p/libccgi/wiki/Home/

Astronomical calculations which can be very calculation heavy and depend on the geographical location of the requester, therefore can't be cached.