Understand that basically all c2 tilemaps are is a sprite sheet without animation. Meaning a series of images. Now Construct 2 tilemaps do not support animated tilemaps on it's own. Meaning sprites are very useful for these animated objects. There are a few work arounds people have come up with other than sprites, but it kills performance all the same. Which can be a drag.
Now when creating a tile map you have to think of widthxheight for each tile. Example: I made one with 64x64 images. Keep in mind if you want to improve performance and have no transparent textures consider using index color mode even if you don't remove colors at all it will greatly improve performance by lower resource cache as the image size will be lower, freeing up more memory. More importantly it removes the need for over redundancy of creating more objects.
Now for collision don't worry about tiles that have transparency within them, the only thing you need to do is setup the collision polygon shape up for them. c2 does not auto setup collision on tiles like it does with sprites. Meaning all you need to do is click on the polygon icon in the tilemap pannel within construct 2. Speaking of the tilemap panel it supports: flipping, mirroring, and 90 degree rotations.
What mirroring, flipping and rotations mean for you is you do not need to create multiples of the same tiles. Unforchently the object tilemaps does not have a angels property, this is a downside, meaning you can not rotate the entire tilemap at a angle based off the player, which means the behavior rotate, and phyics will not work how it is intended. Physics will need to be applied to it how ever if you have objects with physics if you want them to collide with you tile maps. You will just have to make them unmovable, and with non-solid tilemaps you will just have to disable them by the instance if you want to repeatedly use the same tilemap for background as foreground.
If your asking yourself about construct 3's polygon button in the tilemap editor panel then simply double click on the tile and it should popup the polygon editor button for collision.
Larger tilemaps will consume more memory so splitting the tilemaps into multiple tile map objects can help. But only when it is necessary. If you want them to have same code and behaviors and values, I suggest applying family grouping to your tilemaps and referencing that in your event sheets.
Tilemaps are not perfect how ever, they lack animation, they act as a singular object rather than a grouping of them when behaviors is applied, and you may need many of them to do the tasks of tiled backgrounds and other object types. But they are rather useful. You won't be making a sonic clone with only them. IE if you apply opacity to a tilemap it will apply it to all, which can be a good thing but also a bad thing. Ex: Windows its bad for, front walls like going inside a building Good use of opacity.
It all depends on usage and you don't need to keep it square thanks to collision polygon mapping.
You wont be making a sonic clone anytime soon as it lacks angles, you see creating a sonic clone would imply you are able to change the players sprite angle based off the surface he is on's angle. Now yea I know sonic was made with tilemapping but construct tilemapping and sonic are different and how collision is programmed in sonic is massively different. Yes I am aware you can make it based off triggers, and other objects but like I said you won't be doing it based off tilemap objects without them.