Gaze data is a well-established indicator of cognitive load levels during human-computer interaction. The design characteristics of virtual learning environments can be a source of extraneous cognitive load, which can, in turn, strongly impact the gaze behavior of learners. The cognitive load experienced by learners can be classified into three components: intrinsic, extraneous and germane. This study aims to systematically investigate the relations between the design of virtual reality environments, gaze data, and cognitive load. Therefore, a randomized control trial (N = 285) using an experimental-control group with a pre-post-test design was conducted in a physics lab course. The two conditions of the experimental and control groups differed only in the design of the e-learning environment, which led to higher extraneous cognitive load in the experimental group, while reports for intrinsic and germane cognitive load did not differ. The eye-tracking results revealed systematically and significantly different gaze behavior between the two groups (medium effect size). The results provide important insights into the design of future virtual reality learning environments, specifically about minimizing the cognitive load and increasing the efficacy of the learning unit.