First, let make things clear.
- 64-bit plug-in modules (DLL) only work with 64 bits applications hosts.
- 32-bit plug-in modules (DLL) only work with 32 bits applications hosts.
64-bit plug-in modules (DLL) do not sound better than 32-bit plug-in modules. This has nothing to do with 64-bit audio processing.
The main difference between both is the amount of memory they can address. In some scenarios (I am thinking of a sampler), a plug-in may want to reference a large amount of samples (many gigabytes), and 32-bit has it's limitations.
Next, still today, many plug-ins don't exist in 64-bit flavor, but most exist in 32-bit flavor (especially old plug-ins), so in terms of compatibility, 32-bit is the winner.
Lots of sequencers and sampler are streaming audio from disk, so memory may not be issue with 32-bit hosts and plug-ins at short term. At long term, with the tendency of having very large samplers, the memory addressing issue may become a real issue.
To answer your question about EastWest plug-ins, 64-bit is likely to get you better performances as you will avoid streaming samples from disk and be able to sample the complete kit from memory.
Bridges: Some bridge applications exists (jBridge, and Steinbern VST bridge) and allow you to run 32 plug-ins in a separate process and bridge the audio inputs/outputs through a 64-bit host, allowing 64-bit hosts to run 32-bit plug-ins. This method has some performance disadvantages as the bridge has to copy the memory from one process to another and that serialization/deserialization process may take precious CPU cycles. It may also introduce compatibility issues, as a layer is added between the host and the plug-in.
More information about memory addressing here (a bit technical):