Performance differences in creation of MLinExpr and constraint addition in Python API

Hello, I have recently noticed performance differences in the creation of MLinExprs and the addition of constraints from MLinExpr.

import timeit
import numpy as np

setup="""
import gurobipy as gb
model = gb.Model()
x = model.addMVar(shape=(100,100), name='x')
y = model.addMVar(shape=(100,100), name='y')"""


stmt1 = "x-y"
stmt2 = "x.reshape(-1) - y.reshape(-1)"

res1 = np.min(timeit.repeat(stmt=stmt1,setup=setup,repeat=1000,number=1))
res2 = np.min(timeit.repeat(stmt=stmt2,setup=setup,repeat=1000,number=1))

print(f'RES 1: {res1:.5f}')
print(f'RES 2: {res2:.5f}')
print(f'RES1 / RES2: {res1/res2:.5f}')

RES 1: 0.01304
RES 2: 0.00095
RES1 / RES2: 13.78600

Creating an MLinExpr from the reshaped MVars results in a ~14x speed up on my machine

import timeit
import numpy as np
setup1="""
import gurobipy as gb
model = gb.Model()
x = model.addMVar(shape=(100,100), name='x')
y = model.addMVar(shape=(100,100), name='y')
lhs = x-y
"""
setup2="""
import gurobipy as gb
model = gb.Model()
x = model.addMVar(shape=(100,100), name='x')
y = model.addMVar(shape=(100,100), name='y')
lhs = x.reshape(-1) - y.reshape(-1)
"""
stmt = "model.addConstr(lhs==0)"

res1 = np.min(timeit.repeat(stmt=stmt,setup=setup1,repeat=1000,number=1))
res2 = np.min(timeit.repeat(stmt=stmt,setup=setup2,repeat=1000,number=1))

print(f'RES 1: {res1}')
print(f'RES 2: {res2}')
print(f'RES1 / RES2: {res1/res2}')

RES 1: 0.02762570000049891
RES 2: 0.006292300000495743
RES1 / RES2: 4.390397787505745

The addition of constraints from the different shaped MLinExprs results in a ~4x speed up on my machine.

If I understand correctly, the constraints added to the model are equivalent. The only difference is that the returned MConstr object now has a shape of 10,000x1 vs 100x100.

I'm just curious why is this the case? Is there any drawback to reshaping the MVars into vectors for constraint addition?  Am I simply missing something obvious?

Thanks in advance,

Dillon