How is that worse? It takes an insignificant amount of processing time to do it the first time and, because anyone who isn't ******ed would cache the results, no more time than simply switching your shield for any subsequent combinations. The only negative result is n^2 hard drive space being consumed. However, considering the total size of the shields folder is fairly small anyway, I'd say it's perfectly viable and no more than an hour of work.

I think any of the above methods are viable enough. If you're concerned about performance, then know that the other 3 methods add 1 additional draw call and texture swap/state change per player, which is something that any other accessory adds. The last method creates no performance impact on the client (Except the obvious fill rate consumption) and has an extremely negligible impact on the server when you first create the image (Which could be offloaded to a simple script running on another server through sockets, by the way). Past that < 0.0001% of the time, everything can be cached, so no performance implications anywhere.