Yes, UMLS is indeed a great resource! I do think one could use UMLS as a core vocabulary but with specific application context. At my company, we use it for information retrieval of clinical trials. NIH itself uses it in several of their services (MedlinePlus, ClinicalTrials.gov, etc)

In the context of EHR data-entry/documentation, using UMLS would not make sense just because of the inherent noisiness/granularity differences that come about due to the integrated view of the world. With terminologies like SNOMED-CT, you run into coverage issues and end up pre/post coordinating lot of knowledge.