"blocks" : 72225, "networkhashps" : 103314719,
"blocks" : 72419, "networkhashps" : 267124100,
"blocks" : 72598, "networkhashps" : 126571030,
Assuming not some error.
someone added 160k kh/s to the network to make up over 60% of the total hashes per second.
shows that someone has a OpenCl implementation of scrypt-jane and a GPU farm at their disposal.
The code I added to yacoind to produce the "networkhashps" info is only an estimate, and is based on spacing between solved blocks over a fairly short period of time. There isn't a way to directly measure hash rate, the best we can do is estimate it. Variability is capable of producing the swing you observed above. From my own data collection process that keeps track of average block spacing, at this time my opinion is that there wasn't a sustained rise in (estimated) hash rate that would be outside the realm of variability.
Now, if the estimated hash rate jumped and then stayed at that level, that would be more interesting and much less likely to be just a variability-induced fluctuation in the way my hash rate estimation code works.
I copied a good portion of the hash rate estimation code from Litecoin. Our (target) block rate is 2.5x faster than Litecoin. Perhaps I should tweak the code to average the hash rate estimate across 2.5x more blocks than Litecoin does to try to smooth out some of these swings so we're at least averaging over the same amount of time. Even Litecoin's hash rate estimate has some pretty wild swings due to variability.