I gave an example of 10% and 10% because of the '20/80' rule. Results of <20% of trades will drive the optimization results...

The difference between what you suggest and I is in 'strictness' of the overshoot.

If it exactly Entry+Penalty, then the optimization will most likely overfit...If it is '<=Entry+overshoot', it is less likely to be so.

Again, that is mostly needed for efficient discovery/prototyping, with a piece of mind that it is sufficiently realistic. (handling orders in live is the next step/ different discussion)

Would really appreciate if negative Penalty would work as described for Entry and TakeProfit.

It will overfit in both cases, since they are technically equivalent. They produce the same trades. The only difference is an offset to Entry.

The overfitting is not due to Penalty, but because you're likely training on random noise when you optimize Entry in this way.