I've replicated this behaviour in London, and you can blame WME...
When WME requests UR data from the server, the request is always made for an area slightly larger than the visible map area, and the server returns however many URs it feels is appropriate for the requested area - this is the point at which Waze does its own suppression of URs in congested areas.
Now, having received this block of UR data, WME goes off and renders it all, including the URs that are completely off-screen. URO+ then does its own sweep of the UR layer, adding custom icons, mickey mouse ears etc. as required to all of the markers defined in WME - offscreen included.
When panning the map, until you release the mouse button, all you're doing is moving around all of these pre-rendered map layers. Pan far enough, and you'll notice some of the layers run out of data - the UR layer just happens to be one of them, on my 1600x1200 screen, if I pan the map as far left or right as I can go within the bounds of the maximised browser window, the UR layer gives up about halfway across the screen. Only once you release the mouse button does WME place a fresh request to the server for the data associated with this new map view. Therefore, during the panning operation, UR markers will come into view which have already been tweaked by URO+.
However, as soon as WME receives the fresh data from the server, all of those existing markers are blown away and replaced by ones associated with this new data, and if the newly served up data happens to have been pre-suppressed by the server then it's likely that some of those old markers that'd been panned into view will end up disappearing again.
This behaviour is completely reproducible even if WME is run in its unadorned, unenhanced, unbelievably plain (how did we ever manage to edit maps like this???) state, proving that it's absolutely nothing to do with URO+.