The SSL session cache is currently set at 1024 entries. Each session remembers a certificate which is decently large. 1024 was just carried over from some internal OpenSSL value and is quite likely the wrong size.

Session cache sizes are a time vs. space tradeoff. We should gather data on resumption rate vs. cache size and try to find the right value.

The session cache currently lives in a base::MRUCache. One possibility is if we histogrammed, for each session which ultimately was used, what was its MRU rank. That distribution can help guide how many resumptions we'd lose if we shrunk the cache by such and such.

If that's hard to measure, we can also histogram resumption rate against cache size field trial. (I'm not sure if we have histograms on resumption rates. If not, we really really should.)