Ya, it definitely appears like the caps are having a negative effect on your frametime spikes and FPS.
Really not sure how to explain that, but it's present. That said, a 1-2ms difference is well below noticeable levels so as long as 999 isn't giving you the same spikes seen in the 350 plot I would personally stick with that. But it's up to you.
Again here, the number discrepancies themselves could easily just be due to the fact there is no benchmark for CSGO and so you are never going to be able to repeat the same exact video test twice. There will always be fluctuations.
But the number of spikes is the issue here, and it certainly seems to be related to the fps_max.
I wonder, have you tried limiting FPS through other means? Something like Nvidia Inspector or MSI Afterburner? I wonder if it would have the same results or if this is something buggy in the engine code.