@evoskuil don't know if this is of interest to you, but I have a piece of code which is doing EC ops on 100k keys. It takes around 15 secs. I added the --enable-endomorphism switch to secp library and the time dropped to 10 secs. I then noticed in Valgrind that 15% of time is spent parsing the public keys from the internal format. I reimplemented the algorithm in pure secp internals (not the public API, but the actual points in jacobian coordinates called secp256k1_scalar and secp256k1_gej) and time dropped to 6.5 secs. That's a pretty drastic improvement.
