(I tried to post this as a possible bug report but must have done something wrong as I never got a confirmation or response - over a week ago now)
I have 2 machines:
Local: i7-6700K 4GHz/32gb/GTX1660Ti/Windows10 64bit Remote: i7-4770K 3.5GHz/32gb/GTX1660/Windows10 64bit
(both have the latest Nvidia drivers)
With crowd-render 2.0 I rendered a 500 frame animation. It took over twice as long as an F12 render & I found 4 or 5 blank frames.
After 2.1 was released I removed 2.0, installed 2.1 and tried again.
I switched to manual load balancing (0.6 for local, 0,4 for remote). Rendering still took as long - up to 19s per frame as before. The balancing seemed to have an effect, but there was a noticeable colour difference between (I assume) the locally-rendered and the remotely-rendered parts.
I also found that frame 2 was blank (I stopped the rendering after this frame).
On closer inspection of the logs (local attached, I couldn't attach the 2nd log) I found that in the local log Frame 0 was processed twice and Frame 3 wasn't listed (the missing frame was frame 2). The remote log seemed to be complete.
In one other test I had a frame with one part completely white.
I couldn't find any other reference to this type of result so maybe I have a system problem?
Phew, quite an answer, thanks very much James. To your points.
Colour differences:
Embarrassing but true, I am using an external HDRI and I didn't pack it into the blend file - I will correct that.
Speed:
Actually I did try to change the crowd-render settings on the local & remote machine to GPU (CUDA), but I got the impression that it sometimes got reset. I'll try again.
Dropouts:
I have a 1Gb wired network connection for the 2 machines - no WiFi. I'll run a few response time checks. I think I noticed a comment somewhere that you don't like IPv6? I keep switching it off but after OS/driver updates it sometimes gets reactivated. Could that have an impact? In the meantime I relasie I can just r-render the missing frames, there aren't many
I'll keep investigating.
John