Does the introduction of a 'Jumbo Switch' create/defragment jumbo packets on the network (when nothing on the network streams >1500bytes)?

[RobertVision]

In short: Does introducing a jumbo frame enabled switch on a network (where no -really zero- nodes broadcasts/streams jumbo frames) create problems.

Long version: We deploy ad-hoc networks for industrial sensors and automation applications where we occasionally need to broadcast/stream jumbo frames. Such projects always ensures that all NICs and switches etc have jumbo frames properly configured etc. Some projects has no need for jumbo frames indeed.

Now, we need to standardize our hardware, meaning that we will be introducing a standardized switch. The debate ongoing is if we must have two standard switches (one standard configured and one jumbo configured). The argument for these two versions is that introducing a jumbo frame enabled switch on a network somehow magically creates jumbo packets. 

Bonus question: Why isn't the port MTU size on managed switches  jumbo by default?

[Reza Sharifi]

Enabling jumbo frame for all devices should not harm anything even when you do not send jumbo frame packets. This is as long as all devices in the path support jumbo frames.

Bonus question: Why isn't the port MTU size on managed switches  jumbo by default?

Because there may be devices that do not support jumbo frames.

HTH

[Joseph W. Doherty · Accepted solution]

Short: For transit traffic not using jumbos, it will have no effect. For the device itself, it may. For example, if the switch is L3, routing protocols will often take advantage of jumbos.

Jumbo's become a problem, when two devices, on shared media, have different MTUs, as the device with a smaller MTU might be unable to accept a larger MTU.

As Reza notes, to avoid such issues, that's why jumbos are off by default. Also keep in mind, there's no standard for jumbo MTU, so even two "jumbo" enabled devices can have a L2 intercommunication issue.