Concise answer: 2.45 is the first version with the bug.
Quick description of the setup I used to test this:
I put ninjam server on my email peecee. I installed REAPER on my two kids peecees. From those two plus my DAW, I connected to the server over my 100Mbps switched network in the house. So bandwidth is not a consideration.
I created a minimal project on each box, two tracks: ReaNINJAM on track 1, JS:Synthesis/tonegenerator on track 2. Track 2 is routed to 1, and master parent send unchecked, just like my typical NINJAM template. I sent two local channels from each machine, each fed from stereo1/2 (the test tone). All meters show what I expected to see. I can post the .rpp if it's useful.
Following are screenshots of perf meter as I ran through and reinstalled older versions of REAPER on all boxes in the test setup:
2.46: my DAW reports CPU 6.3-40.6%. Crackling makes it unbearable to listen to the signal.
2.45 my DAW reports CPU 6.3-37.5%. Crackling makes it unbearable to listen to the signal.
2.44 it's cured. But I had to get the last 2.45 machine out of the room first. my DAW reports CPU 3.9-22.7%, with the ReaNINJAM track consistently reporting below 1.5%.
The FX CPU number is markedly lower in 2.44. But all machines in the room have to be on 2.44 for that to be true. The screenshots don't show maximum values for FX CPU in the cases where it's high. That number was flying all over the place.
Justin, thanks again for looking at this. I'm very impressed with how easy it was to put up a test lab full of reapers--I only spent 45 minutes or so getting this data together thanks to how quick this program is to install and download.