While much of the virtual-machine (VM) instruction-dispatch overhead has been eliminated in Gforth, it still performs one VM instruction-pointer update per executed VM instruction. The present work optimizes much of that overhead away and reports and analyses the performance effects. Some benchmarks see speedups by factors $>2$ on recent high-performance cores, while other benchmarks and other cores see more modest results, often in the speedup ranges 1.1–1.3. The explanation of the big speedups is that, without the optimization, VM instruction-pointer updates form the critical path in benchmarks with long-running loops. We expect that the benefits of these optimizations also apply to other VMs with light-weight VM instructions.