Hi,
This is correct behavior. If there is an alternate pronunciation in the dictionary that means that there is a common accent that uses it, so if you just include the ones which correspond to your own accent, other users will be excluded. It shouldn’t lead to any unwanted behavior. All recognitions of the various pronunciations will be returned as just the word itself in the hypothesis without the (2) or (3) in the word, since that part is managed by the grammar or language model (which only has the one textual representation) rather than the pronunciation dictionary.