I think that a description of a dragon that says it weight five tons - whether that is in a paragraph of text, or broken out into a "stat block" format, or whatever - is not the same as the sort of stat block we see in a wargame that actually feeds into a set of rules procedures and resolution mechanics. It's just a description.
Will a five ton dragon sitting on top of a house collapse the roof/building? How strong is the roof, and how much load can it take? Answering those questions doesn't seem like a
simulation - it's just doing some engineering calculations. My guess is that, at most tables, people would just make something up.
Do we need a rules system to determine whether or not creatures destroy buildings that they land on - say, a system that takes some sort of Size statistic, feeds in an Architectural Strength statistic, and generates an outcome? Rolemaster has various rules in this neighbourhood, but most people find RM too rules heavy! Which is to say, they don't like simulation-type mechanics and prefer to make things up.
A D&D example: the Wilderness Survival Guide is full of charts and tables that allow determining things like how much food someone might find if they forage in such-and-such an area, with such-and-such a skill, for such-and-such a time. I doubt those tables have seen much use, relative to the whole extent of D&D gameplay in which they might have been used. Again, because many people just prefer to make things up than filter it through a system that models (or purports to model) the process in question.
I basically agree with
@chaochou's reply, but want to add something.
I don't think the gameplay principles are the same. The rules for a simulationist wargame, or the rules for jumping chasms in a system like Rolemaster or RuneQuest, have correctness conditions: we can see whether or not the rules, when applied, produce a spread of results that more-or-less conforms to the real world outcomes that our system is meant to produce. If they don't, we can improve our rules (eg by adjusting paramters, or making room for further inputs, or whatever). At least in my play of RM, this sort of correction-by-reference-to-real-world-constraints is a meaningful part of play.
But in the case of invented units and hypothetical engagements, there are no correctness conditions. The "expected outcomes" are just imaginative preferences. If the 4th level fighter (a Hero, in the Chainmail/classic D&D parlance) is able to handily defeat a 4 HD Ogre in a typical engagement, does the mean I have to adjust my inputs and parameters, or is that the system working as intended? There's no answer other than one's aesthetic preference for Hero vs Ogre fights!
Further, as
@chaochou mentioned, to satisfy correctness conditions for wargames we need
lots of different rulesets even when we make relatively small variations: eg a system used to resolve clashes of Napoleonic infantry, cavalry and artillery will probably not be usable, without significant changes, for a conflict featuring tanks, machine guns or modern artillery.
But D&D uses the same system to resolve a fight between two mercenaries, a fight between one mercenary and five others who surround them, a Hero vs an Ogre, a Hero vs a Dragon, etc. What's it simulating? It's a mechanical process for answering some questions, like
Who wins the fight?, but it's not simulating anything, even hypothetically.