Itīs all about probabilities.
When dispatching with an MEL item you typically have reduced redundancy in a particular system. To take a further and subsequent failure into account the MEL can prescribe a certain limitation to contain that risk.
When something fails in-flight the MEL does not apply as explained very well by some of you. Statistically it is less likely that you suffer a subsequent failure within the same system in-flight, hence this is not taken into account and no limitation applies.
Itīs the same concept with Alert Height. Once below AH a single failure in the aircraft system does not cause a downgrade in the aircraftīs landing capability. The AH in the realm of auto land is analogous to the point where the aircraft starts to move under its own power in the realm of MEL (!)
Vilas has a very good explanation, I would add one important thing:
If I suffer any failure before take-off I will always consider the performance implications and not blindly dismiss the MEL. Any failure affecting the braking/stopping capability of the aircraft has to be addressed! FCOM/ECAM does not address these things. Also in the Airbus in particular you have to be careful that Config 1+F doesnīt become Config 1 due to ADR problems etc....