Using invalid number of parameters with AS, IF, etc. definitely should be a syntax error that fails execution, only reporting such errors at parsing time is not an option. To keep the usage of the parsing model easy and backwards compatible, all these errors should get available via errors attribute of the nodes in the model. Possibly it could be a property or validate() could add errors from tokens there. Either way, it must not have adverse effect for parsing performance.

Because it's unlikely these are common errors and because they are already now detected by the parser, I don't consider it too high priority to detect them earlier. I understand it's a bit better for editors to be able to highlight only the latter value in cases like

IF    $foo    $bar

instead of highlighting the whole IF node, but editors anyway need to use the latter approach in cases like

IF

where the problem is that a token is missing. I would assume it would be easier for editors to just check does the IF node have errors instead of going through each tokens. At least for Robot itself the current validation approach is simple and I'm not at all certain making it more complicated is worth the benefits. Enhancing parser to detect error that currently are only detected at the execution time (#4210) would have more benefits and would probably also be easier.