Many interfaces exist for locomotion in virtual reality, although they are rarely considered fully natural. Past research has found that using such interfaces places cognitive demands on the user, with unnatural actions and concurrent tasks competing for finite cognitive resources. Notably, using semi-natural interfaces leads to poor performance on concurrent tasks requiring spatial working memory. This paper presents an adaptive system designed to track a user's concurrent cognitive task load and adjust interface parameters accordingly, varying the extent to which movement is fully natural. A fuzzy inference system is described and the results of an initial validation study are presented. Users of this adaptive interface demonstrated better performance than users of a baseline interface on several movement metrics, indicating that the adaptive interface helped users manage the demands of concurrent spatial tasks in a virtual environment. However, participants experienced some unexpected difficulties when faced with a concurrent verbal task.