The above materials should help you to work through managing your point clouds - the Variable Density Modeling is likely the most important information relating to your question, however used in combination with the Sampling tools you can likely improve it further.
The Sampling tools offer two methods today to sample a Point Cloud Region
The Spatial Algorithm is trying to find a near point to your Spatial Sampling parameter - ie if you say every 2' then it is looking for the nearest point to your 2' pacing.
The Random sampling if you say you want 1000000 points out of a 10,000,000 Point Point Cloud then it will effectively extract the 10th point from 10 points based on order of measurement.
Neither method is looking at the shape of the "surface" as it extracts the point data
My method of working is as follows
1) Get your scans into the Project and aggregate them into a Point Cloud Region called e.g. August Survey
2) Bound the Point Cloud to the area that you are working in and select all the data outside and move it to a new point cloud region called August - Not Needed or similar
3) You can use the Classify Regions or Extract Ground functions to try and eliminate Tree / Shrub canopy and to try to break out Buildings, Poles, Vegetation etc.
4) Use the Polygon Select method combined with Limit Box to select and move Bad Data eg Vehicles, People, Noise, Piles of pipe, and other things you dont need into the Not Needed Point Cloud Region
5) After all those steps you are left with the Point Data that you will use to form your surfaces.
6) Within the remaining data, you may want to create an extra Point Cloud Region, and move data into that selected by Polygon Select method that you need to model at Higher Density than other areas that you will be happy with lower density - you can create regions of Low, Medium and High density need as you see fit
7) Set your Support - Options - Point Clouds settings for the number of points in a surface - refer to the exercise details that shows how Variable Density Modeling can be used (this is the secret in BC-HCE to getting increased Fidelity where you need it and Lower Fidelity where you dont, and trying to keep your surfaces as small as possible but as accurate as you need them.
8) Build up your surface by adding Point Cloud regions in incremental steps - remember that each Region you add after the surface is first created will extract however many points you have specified in Support - Options - Point Clouds - and in this way you can change the density of points. It is sometimes worth looking at the number of points in each of your regions to determine what you think may be required to create the model that you want to create.
With experimentation, I have found this to work really well and I can typically get what the end user wants in a surface model using these methods. Having said that I am typically working with Civil Earthmovers and Rock Excavation teams and they have different needs to a pure land surveyor
You can also look at the Feature Extract methods if you need to pull out feature lines to assist forming good surfaces along Features like Flow Lines or top of curb etc.
Polygon select in Plan or 3D view (3D especially) is a great tool to islolate areas like curb or barier or retaining wall that you want to model at higher density than areas like the Pavement or embankment areas
Hope that this helps
Alan