@Richard Hernandez good question and honestly a really important one to nail down before you lock in a topology.
The 2Mbps spec is the ceiling for the module, but at 3-4km ground-to-ground you're going to be working with whatever PHY gives you the most reliable link, not the fastest one. At those distances with both nodes on the ground, the module is doing its work through the external amplifier and the receiver sensitivity tuning — and that's where the range comes from, not from burning bandwidth. Realistically at 3-4km you'd be on 1M PHY or potentially coded PHY depending on your environment, and you should plan your budget around that rather than 2M.
Here's the thing though — for your use case it genuinely doesn't matter. You said 200-400 bytes every few seconds per node. Even coded PHY S=8 gives you something in the neighborhood of 125Kbps effective, and your burst cadence is nowhere near saturating that. A 400 byte packet at 1M PHY with BLE overhead is done in well under a millisecond of airtime. The PHY fallback isn't your bottleneck here at all — latency and connection reliability are what you actually need to validate.
What I'd suggest is getting a pair of modules and testing your actual RSSI floor at your worst-case node spacing before committing the mesh topology. The 5km ground spec is real but that's line-of-sight with decent antenna placement — if you've got terrain, vegetation, or structures in the path at those distances, that's where the margin gets eaten. The throughput budget you're describing should be totally fine, but the link reliability under your specific site conditions is what's worth stress testing early.
