Computing a query plan is a very complex process that relies on systemic. There is many conditions to obtain a query plan that will be exactly the same on two similar machines... The most important of them are:

• the resources of the different machines in terms of hardware (CPU, RAM, disk organisation and speed...)
• the fact that statistics is generally created/updated by a random sample

Other conditions can increase the differences like:

• the usage of a VM
• the defragmentation of indexes and tables

Are you sure that thoses things are strictly the same ?