Sample Set size requirement - Boosted Decision Trees (gbm package) Models

Is there any formal literature or generally accepted "Best Practices" for the sample set size needed to build a Boosted regression trees?

I have a client with only 188 records and my only way so far to validate the results is to continuously test different subsample amounts of records and measure performance against the holdout records. The smallest model build was with 101 total records.

All of my results so far have been similar - accuracy ranges in 78 - 81%. I figured if the accuracy swings were greater than there'd be a reason to question the model results. As is, I feel somewhat comfortable.

Any thoughts?