There are already many good answers here that address your question from a hypothetical point of view. I want to add to this discussion some empirical data coming from AI chess, based on the arXiv preprint Assessing Game Balance with AlphaZero: Exploring Alternative Rule Sets in Chess (2020) by Nenad Tomašev, Ulrich Paquet, Demis Hassabis and Vladimir Kramnik (link: https://arxiv.org/abs/2009.04374). The first three authors are from Google's DeepMind team that developed the AlphaZero chess engine, and Vldimir Kramnik is of course a former chess world champion.
The paper arose from a similar question: what chess would look like without castling, and whether this would lead to more interseting and/or decisive games. Since the AlphaZero chess engine is entirely based on self-learning, it can easily be used to address such questions: simply change the rule set in the program, and let it train against itself. After enough training, it will out-perform any human player on that chess variant.
The team uses this approach to compare classical chess with nine variants, including the variant where forcing stalemate is a win. As you can see in the paper, they analyse each variation in quite some depth. Your question about decisiveness of games under this rule change is addressed in Figure 2: in 10,000 games at 1 second per move, 10% ended in white victory and 86% in a draw, compared to 7.8% and 88% respecively for classical chess. In 1,000 games at 1 minute per move, 2.5% resulted in a victory for white and 97% in draw (classical: 1.8% and 98% respecively). So indeed, your proposed rule change leads to slightly more decisive games, but the difference is quite small.
Of course, this is computer chess, and human chess might be different.