@Raghu Paripati If you are looking to train a model and would like to know the limit for number of documents that are required to train, then the guidance is to use atleast 5 or more documents of good quality to train the model. All the requirements are mentioned in the documentation here.
The number of documents that can be passed to the model for extraction is basically limited by the pricing tier of your form recognizer resource and the document type. As per the pricing page if you are on free tier you could only pass 500 pages but for other S0 tiers the charge is 50$/1000 pages for custom and invoice document types and 10$/1000 pages for all pre-built models.
In case of custom models when the analyze POST call is made the request is accepted and the result is available via a GET call to read the result. This makes the service asynchronous and you should not have any problems with the performance too. I hope this helps.