Are gdb1 and gdb2 completely different DBMS instances or databases? And if you reverse the order, gdb2 is the slow one and gdb1 the fast one?
I suspect there is either a library or collection of libraries that get loaded by the first Make Feature Layer call involving an enterprise geodatabase. Since the libraries are already loaded, subsequent enterprise geodatabase connections go much faster, even connections to other databases.
What if you do something like:
ws2 = r'\\somenetworkpath\gdb2.sde'
arcpy.env.workspace = ws2
arcpy.ListFeatureClasses()
ws1 = r'\\somenetworkpath\gdb1.sde'
fcname = r'gdb1.someschema.FCNAME
input_fc1 = os.path.join(ws1, fcname)
arcpy.MakeFeatureLayer_management(input_fc1, "input_fl1")
ws2 = r'\\somenetworkpath\gdb2.sde'
fcname = r'gdb2.someschema.FCNAME
input_fc2 = os.path.join(ws2, fcname)
arcpy.MakeFeatureLayer_management(input_fc2, "input_fl2")
or
tmp_fc = arcpy.CreateFeatureClass_management("in_memory", "tmp", "POLYGON")
arcpy.MakeFeatureLayer_management(tmp_fc, "tmp_lyr")
ws1 = r'\\somenetworkpath\gdb1.sde'
fcname = r'gdb1.someschema.FCNAME
input_fc1 = os.path.join(ws1, fcname)
arcpy.MakeFeatureLayer_management(input_fc1, "input_fl1")
ws2 = r'\\somenetworkpath\gdb2.sde'
fcname = r'gdb2.someschema.FCNAME
input_fc2 = os.path.join(ws2, fcname)
arcpy.MakeFeatureLayer_management(input_fc2, "input_fl2")