This is a great question that's difficult to answer. There are differing opinions out there regarding when and to what-level to perform hydrologic conditioning, but for our purposes, the solution lies in exactly what question you're asking of the dataset. I strongly recommend calculating the hydrologically-dependant terrain attributes only if you're calculating them with an end-purpose, project, or hypothesis in-mind. This will help guide the steps you take along the way in performing your terrain analysis, and create products that are the most useful for the specific applications you're planning.
As you touched on, there are many levels of hydrologic conditioning that can be performed, from removal of digital dams, to creating models which take into account soil types, run-off curves, specific rainfall events (i.e. 100-year), and the like. While many of these decisions are budget-driven, again the question comes down to what do you need the data to do for you.
For example, much of the work we do involves targeting conservation efforts to portions of a watershed or county that need soil-erosion or water-quality remediation. A LiDAR Terrain Analysis in this instance is used to deliver a 1st cut of potential sites in which to invest conservation dollars. Hydrologic conditioning here, while useful, may not be financially feasible or absolutely necessary to find the largest contributors of erosive overland flow. Of course, a complete accounting of these features could always be aided by hydrologic conditioning, but there are no hard or fast rules to determine where on the sliding scale of data-quality you'll need to be.
Contrast that scenario, with an engineer who may be using terrain attributes to determine a flow budget for a particular stretch of stream, complete with a fully-derived stream network and values that are tied directly to monitoring data. In this instance, the best numbers will most definitely be achieved by a thorough hydrologic conditioning of the DEM prior to running various terrain attributes.
Landscape factors also come into play, which can make matters more difficult. Highly deranged and human-altered drainage (Red River Valley, South Central MN, etc.) can drastically affect water movement on the landscape, making hyrdologic conditioning more useful in interpreting the resulting attributes.
Feel free to consult the materials from some lectures posted here - http://wrc.umn.edu/randpe/agandwq/tsp/lidar/LiDARTrainingMaterials/index.htm.
They're part of a series of workshops we've done on LiDAR, Hydrologic Applications, and Terrain Analysis, and have a few more examples that can help you in making your decision.
Good luck, and let us know if you have any other questions.