It isn't a long way from reliability slider to engine tweaking, you just set the power output/reliability and hope it will last through the race. Like in the other end you have Renault F1 and in the other one McLaren F1. Other is fast but unreliable and the other one not so fast but reliable.
As to the whole thing about mechanical failures and its randomness, almost all mechanical gadgets, for example pistons, crankshafts etc. are engineered by using some probability functions and graphs. The variables are: frequency (if moving part), forces affecting to the part, part material characteristics and the form of the part itself. Different materials can handle varying loads very differently.
There are some engineering methods like:
safe life (part won't break down in certain time in certain conditions),
fail safe (the system is allowed to break at some points but the system is still operating) and
damage tolerance (certain parts are changed at times to prevent any damage coming critical in the system). Why am I speaking about engineering methods...read on->
So what kind of engineering methods are used in building the lfs cars? Do we want to inspect the car at times to ensure it is OK, or do we just change everything between the races? Real race cars usually get put into pieces between the races to check and to change some of the parts. And some of the parts may have some errors in it that we can't really see and be sure. Here we use some probability graphs and equations to check if the part
probably will last the next event. So they are using probability functions! So actually the mechanical probs are random to some extend.
If we don't want to completely model the physics related to part mechanical failures we must use some kind of probability equations. The variables in these functions cover the estimated part life (how much stress the part was engineered to handle, i.e. the resistance to break down), forces affecting to the part (rpm, load, heat...) and a memory that keeps track of the usage of the part (part condition in general).
So the "mechanical failure system" makes all the time some adjustments to the part's condition and the part will break sooner or later. A good system that does all this has to handle a lot data as there are many many parts in a single engine that can break.
The guys developing the "Racing Legends" simulator threatened to make a sim where all damage is physically modelled. The problem is that we can't make 100% sure damage modelling even in the best commercial "physics simulators" (like Femlab and such). So the only way to model the mechanical failures is to use probability functions where there are some factors we can have some effect (like clutch usage and driving line (over the big curbs or not...).
If some one even reads all this
EDIT. I just red this and it seems that I forgot the point there .
So what I was implying was that if we want to have as realistic damage modelling as possible it will be random to some extend. The randomness derives from the parts and their little deviations (cracks, material properties etc...) so each part is little different and acts little differently. Also there are human errors as some one said earlier which are impossible to predict if we had them in lfs. Btw. should some cars be more reliable than others? What about damage that couldn't be repaired? Should those human errors be in lfs? (even if they were totally random, bad luck, as you can have no influence to these)
Yeah, point made