No, I do not save the number of CPU hours required to sieve each number, but it would not be difficult to obtain a rough estimate. As an example, with about 2000 active hosts working on it, a number with SNFS difficulty of 245 takes roughly about three days to sieve. Considering that many hosts are running multiple projects or only part time, it probably takes around 75,000 CPU-hours, give or take a factor of 2.

I do save the parameters for each run. For example, for 5,377+ (with SNFS difficulty of 243.2), using a sieve region of 32768 x 16384, a factor base bound of 80 million and a large prime bound of 2^30 on each side, and sieving on the rational side, about 3.59 million special-q from the range 45-110 million were sieved, yielding 97.8 million unique relations. Filtering of these relations produced a matrix with approximately 9.25 million entries on each side with a density of approximately 70 non-zero entries per column.