First performance consideration with live views is that, naturally, live view functionality has a price. Maintaining the view in sync with base data is fast and optimized, but still it consumes some resources when base data changes and the view is maintained. And it consumes some additional resources when it is first populated as well. That additional cost is not high, but it exists, manifesting itself mostly in additional memory consumption: when a live view is populated, it creates some internal data in memory to help it maintain itself faster on base data changes.
This additional memory consumption is moderate—roughly equivalent to the amount of memory needed for the resulting list itself. So, although additional resources needed for live view functionality are light to moderate, the obvious suggestion is to use live views only where you are really interested in keeping the result of a query live, or, in other words, where you need that result more than once and the base data is changing so that the result is changing too.
Other than this, there are two different aspects to live view performance:
Query performance is how long it takes to populate the view for the first time by executing the query (or re-populate by re-executing the query, see View.Rebuild). This is covered in LiveLinq query performance: logical optimization and LiveLinq Query Performance: Tuning Indexing Performance.
Maintaining live views on base data changes is optimized using techniques known as Incremental View Maintenance. It means that a view does not simply re-populate itself, it adjusts to accommodate base data changes in a more smart way. The changes to the view are made locally, incrementally, calculating the delta in the view from the delta in the base data. In most cases, it is very fast and does not require any special performance tuning from the user.
As a guideline to estimating the time of view maintenance, we can say that it is roughly proportional to the number of changed items (including added and deleted ones) in the view's resulting collection. So, if only few items change as a result of a change in base data, the view maintenance time is very small and probably negligible, but it can be considerable if a considerable part of the view result changes. It can even become more costly to maintain a view than to re-query it from scratch, if, for example, half of the resulting list changes.
Finally, it is worth noting that maintenance performance for a view does not depend on its query performance. It does not matter how long it takes to populate the view initially; maintenance time is roughly the same and depends on the size of the change in (delta) and not on the size of the overall data.
