The idea I had for this is to multiply everything in a recipe (so more things are required for intake, which produce more items/gases/fluids, but take longer to craft), until a multiple of 100% is the loss chance for every metal involved. Then, multiply the input metal with its percentage and set it in the recipe, 100% requiring one plate, 200% requiring two plates, etc. Example adapted to one of the recipes in the images:

Petroleum gas
input: 1 iron plate, 1 copper plate, 100 methane, 100 steam, in 2 seconds
(0.5; 0.5; 50; 50 /seconds respectively)
output: 80% iron plate recovery chance, 60% copper plate recovery chance, 1 heavy, 5 light oil, 20 petroleum gas, in 2 seconds
(0.4; 0.3; 0.5; 2.5; 10 /second, respectively)
net: -0.5+0.4=-0.1 iron; -0.5+0.3=-0.2 copper; -50 CH4; -50 steam; +0.5 heavy oil; +2.5 light oil; +10 petroleum gas /second, respectively

The modified recipe should look something like this:
input: 1 iron plate (5 comes in, 4 goes out, 1 is consumed), 2 copper plates (5 comes in, 3 goes out, 2 are consumed), 500 methane, 500 steam, in 10 seconds
(0.5-0.4=0.1; 0.5-0.3=0.2; 50; 50 /second respectively)
output: no plates, 5 heavy oil, 25 light oil, 100 petroleum gas, in 10 seconds
(0.5; 2.5; 10 /second, respectively)
net: -0.1 iron; -0.2 copper; -50 CH4; -50 steam; +0.5 heavy oil; +2.5 light oil; +10 petroleum gas /second, respectively

The two net results of the recipes are identical, thus the change wouldn't impact anything other than the feedback loops. The long oil refining times and relatively high outputs of the recipes are more in line with vanilla values, should you be trying to be vanilla-like.

I understand that the idea is to start with some metal and have some left over, but copper and iron is always in enough abundance to make this "catalyst" behaviour an unnecessary complication, which impacts setup tileability very negatively, having to prioritise leftover iron and copper as input, lest the system backs up and cannot produce valuable hydrocarbons. If someone would choose to try this, they would not have to touch existing setups, either.

Hey. Thanks for the feedback. I'm aware this is an issue and I'm working on it from a different angle, currently: I have filed a feature suggestion. It's invisible to you, but these recipe ingredients and recipe results are flagged with "catalyst_amount" values, and I've made a feature request that the engine return all results with a matching catalyst_amount ingredient back to the input chest of the assembling-machine. If this doesn't get implemented, then I agree I will need to do something like you suggest, because handling the probability is annoying when using logistics. To do it properly requires some filtered logistics. I should post a picture of how I do it. Hopefully this will be resolved soon. Thanks again.

However, I may upload an update in a moment to do a quick-fix recipe, pending any base-game changes, just so it's not so annoying.

Updated. I felt since coal liquefaction only takes 5 seconds, Fischer-Tropsch could still be the same. Really what the change does is scale methane/natural gas inputs to 500, with results that are more on the order of the vanilla oil products that are scaled to 5-second processing time. So I think this is a positive improvement all around that just makes natural gas and methane feel less dense, which they are!