Unfortunately, I can't understand the point about random updates of the image blocks and why this affects the delay.

When you tried with a single image block, you only reduced the "schema" requested on the client side to one field (the upper left block, for instance), is it right? And didn't you get any latency improvement?

Commenting this from Lightstreamer point of view, if we rule out client-originated overhead, it remains to check for server-originated overhead, though we don't expect it to be higher. Please, furtherly reduce the load by only pass a single image block (the same block as in the client schema) to the Update method call on the Remote Data Adapter side and see what happens.
If the delay is still high, we can move to Server log analysis (which is difficult now, when all image blocks are involved).