OK, so it looks like it's now only using n=physical cores instead of n=virtual cores. That explains both the decrease in CPU time and the increase in elapsed time - including the virtual cores is measured as doubling the CPU time, but in reality only gives about 25% performance increase - which aligns nicely with your elapsed time results.
Personally this now doesn't worry me: I already use the option to use only half the cores found, as I'd rather take the 25% time hit to keep the system more usable while it's solving. But I agree that the option should ideally remain, for those who want to absolutely minimise the solve time.