In short
Complexity is mastered by breaking complex matters down into simpler ones. Based on this principe, an UML model of a complex system is made of many simpler diagrams. So keep each of it as simple as possible but not more.
Golden modeling rule
Independently of UML, there is a limit to graphical visualisation: any diagram with more than 6-7 elements will appear difficult to understand to the average reader. Keep your diagrams below this limit.
Urban legend? No! Experiments on short term memory by AI pioneer Herbert Simon proved in the 70s this limitation of the human brain. Modeling experts can easily read more complex diagrams, but only because their experience trained them to cluster items, and with this trick they can go up to 6-7 clusters.
Still not convinced?
A sequence diagram only show one specific scenario under the viewpoint of interactions (message exchanges) between objects/parts inside the system.
For alternative scenarios in the same diagram, combined fragments are needed. Your diagram will grow very big, have nested messages, and might quickly become unreadable. I let you extrapolate what might happen if you'd add all the scenarios of all the use-cases, with all the objects...
Last but not least, interactions are not always the best choice to explain what happens. Activity diagrams or state diagrams are very interesting alternatives in many situations.