CISCO - how do jumbo frames work on point to point links? Fragmentation?

Quick question for you all,

I'm wondering how to properly implement Jumbo on my network.My team is looking to start off by implementing Jumbo frames on our Cisco L3 switches from our access layer to our distribution/core layers on point-to-point links.

My question is:

Say I have 5 x L3 Vlans (with respective L2 vlan mapping) where all switch ports are configured for the standard 1500 MTU.I then configure another SVI Point to point link to my distribution layer, both sides configured for JUMBO MTU (Let's say 9000).

Would fragmentation not occur from the distribution layer to access layer during the conversion from Jumbo MTU to standard 1500 to reach my access vlan's?If not, how exactly do the switches perform this?

By doing this, am i just increasing CPU use due to fragmentation with no real traffic optimization benefit?

Thanks